Watts Up

Private AI that runs anywhere

Frontier AI — No GPU Required

We're building AI inference and training engines that run frontier-capable models on commodity CPUs. No GPU. No high memory. Fully local and private. Our architecture combines sparse expert routing, linear-complexity attention, and extreme quantization to eliminate the hardware barriers to AI.

Built from scratch in Rust and C++. Tested at 130B+ parameters.

Any CPU

Trains and infers on commodity hardware. No GPU or specialized accelerator needed.

🔒

Fully Private

Runs locally. Your data never leaves your machine. No cloud dependency.

📈

Scales Free

Performance stays constant as model size grows. Add capacity without adding cost.

How It Works

Sparse Expert Routing

Millions of tiny experts, only a handful active per token. Compute per token is constant regardless of total model size.

Linear Attention

Recurrent inference with fixed-size state — no KV cache, no quadratic scaling. Chunkwise parallel training for hardware efficiency.

Extreme Quantization

1.58-bit ternary weights with low-bit activations. A 130B model fits where a 7B model normally would.

Raw Byte Processing

No tokenizer. The model learns directly from raw bytes, eliminating vocabulary bottlenecks and enabling true multimodal input.

Also: Deep Systems Engineering

We also take on consulting and contract work across our areas of expertise.

Rust Systems Development

SIMD/AVX-512 optimization, distributed architectures, mTLS/PKI, CUDA interop. Microkernel OS, container runtime, and orchestrator — all from scratch.

Infrastructure & Security

10,000+ server fleet hardening. Terraform/OpenTofu on AWS. Built replacements for Docker, Kubernetes, GitHub, PostgreSQL, and S3.

Bitcoin & Crypto Infrastructure

Mining software, SHA-256 structural analysis, block construction, wallet infrastructure. Novel algebraic research on proof-of-work optimization.

Financial Operations

Controller-level accounting, ERP implementations, multi-entity close, process automation. Month-end close in under five minutes.

Technical Stack

RustC/C++PythonCUDA AVX-512 / SIMDeBPF / XDP Mixture of ExpertsRetNetBitNetByte-Level Models ML Training EnginesML Inference Engines Microkernel OSContainer RuntimesOrchestration Post-Quantum CryptoMesh VPNGossip Protocols SHA-256 InternalsBitcoin Core mTLS / PKITerraform / AWSLinux Hardening 10K+ Server FleetsGit InternalsPostgreSQL Internals

Who We Are

Watts Up Trading LLC is Beau and Chance Watts — a two-person team in Houston, TX.

Beau Watts

Systems engineer. Background in high-stakes operations — healthcare, international logistics, financial management — before moving into systems programming and AI. Builds things from first principles, usually from scratch, usually under pressure.

Chance Watts

Operations. A decade of retail leadership and a trading business built from the ground up. Keeps the business running and makes sure things actually ship.

Get In Touch

Interested in our AI engine, need consulting, or want to collaborate? Reach out.